AIML Solution for:​ Sketch-to-face Matching

Match to Catch! This is the concept of our solution for sketch-to-face matching to narrow and filter down, reducing effects of human errors (ie poor sketch quality) and better prediction of suspects.

Comment

Inspiration

Our inspiration came from poor quality / inaccurate sketches that are provided to law enforcers to catch the suspect. (See this: https://www.youtube.com/watch?v=8Db3rluT76w )

What it does

Our solution takes in a sketch and attempts to match (based on KNN) photo provided in the database after performing feature extraction. As such, with a database of photos of suspects, we are able to predict the likely suspects even with a rudimentary sketch.

How we built it

1) Pre-processing of images ​ Dataset: CUHK CUFS dataset​ What we did for the pre-processing includes:​ (a) Gray-scaling ​ (b) Augmentation ( achieve by rotating, translating, shearing) 2) Feature extraction of images​ SIFT Algorithm:​ Detect key points and descriptors​ Extract important information (ie facial features, shape of the head etc ) 3) K Nearest Neighbor algorithm (KNN)​ Concept: Similar data points exist in close proximity​ Classify image by finding distances between a query (sketch) and examples(photos) in the dataset

Challenges we ran into

Beginner in ML, hence very unfamiliar with the different methodology and ML libraries Inexperience in collaboration when in comes to developing a program as a team

Accomplishments that we're proud of

We manage to come out with a product within a short period of time We put in maximum effort

What we learned

Greater understanding about AIML Different ML methodology Image processing Image Augmentation

What's next for AIML Solution for:​ Sketch-to-face Matching

Increase accuracy for our model Better dataset to train our model

Built With

Share this project:

Updates